Improve maintainability of parsing tests

[jclark]

The parsing tests should be designed in a way that when the AST data structure is changed, it is only a little work to get the tests working.

[jclark]

Some thoughts on how to do this:

• expected output should be in .json files, so they can easily be edited
• the AST should be transformed into a JSON format: the tests should work by comparing the expected and actual results in this format
• the JSON format should not be the same as the AST format; the AST format has design decisions that may change; instead it should be a neutral tuple-based format e.g. an addition can be represented by ["+", L, R].
• positions should not be in the JSON format: it is not practical to verify that they are correct by hand; alternative ways should be found to check position information
  • first of all check that the tokenizer is giving the right positions (since tokens are literal strings, this should be easy)
  • then check that the position of each node in the AST corresponds to a token of the right type (i.e. an addition should correspond to a "+" token)

[jclark]

Note that with this approach, parsing tests can be shared with jBallerina.